Marine peptide activity prediction method based on deep learning

Abstract

The present invention relates to the field of data processing technology, and in particular to a method for predicting the activity of marine peptides based on deep learning. First, marine peptide sequence data are collected and preprocessed, and feature encoding and feature fusion processing are performed on the preprocessed marine peptide sequence data to obtain a comprehensive feature set of the marine peptide sequence data; then, a marine peptide activity prediction model is constructed and trained, and the trained marine peptide activity prediction model is used to predict the activity of marine peptides based on the comprehensive feature set of the marine peptide sequence data. The technical problems of inaccurate data processing and poor prediction accuracy in the process of predicting the activity of marine peptides are solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for predicting the activity of marine peptides based on deep learning. Background Art

[0002] As an important component of marine bioactive substances, marine peptides have shown great application potential in the field of biomedicine. Marine peptides have a variety of biological activities, such as antibacterial, antiviral, antitumor and immunomodulatory, and thus have become an important resource for the research and development of new drugs. However, the study of the biological activity of marine peptides is complex and challenging. The main reason is that marine peptides are of various types and complex structures, and traditional experimental research methods are time-consuming, labor-intensive and costly. With the development of artificial intelligence and bioinformatics, computational prediction methods have gradually become an effective way to study the activity of marine peptides.

[0003] In the prior art, methods based on machine learning and deep learning have achieved certain results in predicting biological activity, but there are still some problems and shortcomings. Traditional machine learning methods rely on manual feature extraction, which easily ignores complex patterns and structural information in sequences, and shows limitations when processing high-dimensional data. Although deep learning methods can automatically extract features and have strong expressive power, they often face problems such as overfitting, high computational complexity, and large data requirements in practical applications. These problems limit the generalization ability and practical application effect of marine peptide activity prediction models to a certain extent.

[0004] However, the above technology has at least the following technical problems: insufficiently accurate data processing during the prediction of marine peptide activity and poor prediction accuracy. Summary of the invention

[0005] The present invention solves the technical problems of inaccurate data processing and poor prediction accuracy in the process of marine peptide activity prediction by providing a marine peptide activity prediction method based on deep learning.

[0006] The deep learning-based marine peptide activity prediction method of the present invention specifically includes the following technical solutions: The deep learning-based marine peptide activity prediction method includes the following steps: S1. Collecting and preprocessing marine peptide sequence data, performing feature encoding and feature fusion processing on the preprocessed marine peptide sequence data to obtain a comprehensive feature set of the marine peptide sequence data; S2. Constructing and training a marine peptide activity prediction model, and using the trained marine peptide activity prediction model to predict the marine peptide activity based on the comprehensive feature set of the marine peptide sequence data.

[0007] Preferably, the S1 specifically includes: performing data cleaning on the acquired marine peptide sequence data, and in the data cleaning step, introducing and calculating the sequence homology score.

[0008] Preferably, the S1 also includes: performing topological sequence feature coding processing and molecular orbital energy level feature coding processing on the preprocessed marine peptide sequence data, respectively, to obtain a topological feature vector of the preprocessed marine peptide sequence and a molecular orbital energy level feature vector of the preprocessed marine peptide sequence.

[0009] Preferably, the S1 also includes: based on the preprocessed topological feature vector of the marine peptide sequence and the preprocessed molecular orbital energy level feature vector of the marine peptide sequence, using a topological-orbital feature fusion algorithm to perform feature fusion to obtain a comprehensive feature set of the marine peptide sequence data.

[0010] Preferably, the S2 specifically includes: introducing an adaptive dynamic feature optimization algorithm during the training process of the marine peptide activity prediction model.

[0011] Preferably, S2 also includes: in the implementation process of the adaptive dynamic feature optimization algorithm, initializing the model parameters of the marine peptide activity prediction model, including the marine peptide activity prediction model parameters, the initial learning rate, the first-order momentum parameter, the second-order momentum parameter and the constant used to prevent division by zero errors; and calculating the gradient of the loss function to the model parameters of the marine peptide activity prediction model.

[0012] Preferably, S2 also includes: in the implementation process of the adaptive dynamic feature optimization algorithm, updating the first-order momentum estimation and the second-order momentum estimation based on the gradient of the loss function to the model parameters of the marine peptide activity prediction model; and correcting the first-order momentum estimation and the second-order momentum estimation.

[0013] Preferably, S2 also includes: in the implementation process of the adaptive dynamic feature optimization algorithm, dynamically adjusting the learning rate based on the revised first-order and second-order momentum estimates; updating the model parameters of the marine peptide activity prediction model through the dynamically adjusted learning rate and the revised first-order and second-order momentum estimates.

[0014] The beneficial effects of the technical solution of the present invention are: 1. Redundant sequences are removed by calculating sequence homology scores to ensure the high quality and validity of marine peptide sequence data; missing values ​​are filled by a sequence interpolation method based on Lagrangian interpolation, which improves the integrity and accuracy of marine peptide sequence activity data and provides a solid foundation for subsequent feature extraction and marine peptide activity prediction model training.

[0015] 2. Through feature coding, the marine peptide sequence data is converted into a comprehensive feature set including topological features and molecular orbital energy level features. The topological sequence feature coding and molecular orbital energy level feature coding not only take into account the physicochemical properties of amino acids, but also combine the structural information and chemical substructure characteristics of marine peptide sequences, thereby comprehensively reflecting the multidimensional characteristics of marine peptides; the topological-orbital feature fusion algorithm is used to effectively fuse the topological feature vector and the molecular orbital energy level feature vector to generate a comprehensive feature vector. By weighing and standardizing the importance of different features, it is ensured that the comprehensive feature set composed of the comprehensive feature vector has higher accuracy and comprehensiveness in representing the activity of marine peptides.

[0016] 3. Introduce an adaptive dynamic feature optimization algorithm to improve the training efficiency and convergence speed of the marine peptide activity prediction model by dynamically adjusting the learning rate and momentum correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the method for predicting marine peptide activity based on deep learning described in the present invention. DETAILED DESCRIPTION

[0018] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0020] The specific scheme of the method for predicting the activity of marine peptides based on deep learning provided by the present invention is described in detail below with reference to the accompanying drawings.

[0021] Refer to the attached Figure 1 , which shows a flow chart of a method for predicting marine peptide activity based on deep learning provided by an embodiment of the present invention, the method comprising the following steps: S1. Collecting and preprocessing marine peptide sequence data, performing feature encoding and feature fusion processing on the preprocessed marine peptide sequence data, and obtaining a comprehensive feature set of the marine peptide sequence data.

[0022] Marine peptide sequence data are obtained by collecting public databases and literature to obtain historical marine peptide sequence data. The collected marine peptide sequence data contains marine peptide sequence information and its related biological activity data. In order to ensure the quality and validity of the marine peptide sequence data, a series of preprocessing operations are performed on the acquired marine peptide sequence data. Specific implementation process: The acquired marine peptide sequence data is cleaned. In the data cleaning stage, in order to remove redundant sequences, sequence homology scores are introduced and calculated. The redundancy is determined by calculating the homology scores between each pair of sequences. The calculation formula is: , It is a marine peptide sequence and marine peptide sequences The homology score is used to measure the similarity between two sequences, with higher values ​​indicating higher similarity; It is a marine peptide sequence The amino acids, It is a marine peptide sequence The amino acids, indicating specific amino acid residues in the marine peptide sequence; is the position index of the amino acid; is the indicator function, when The value is 1 when the two amino acids are identical, otherwise the value is 0. is the marine peptide sequence length, which indicates the total number of amino acids in the marine peptide sequence; Used to introduce nonlinear transformations to represent marine peptide sequences Middle The sine and cosine values ​​of the positions of the amino acids are introduced to help capture the complex patterns in the sequences of marine peptides. It is a logarithmic function used to adjust the scale of the homology score and reduce the impact of the length of the marine peptide sequence on the calculation results.

[0023] The similarity threshold is set by expert experience. If the marine peptide sequence and marine peptide sequences The homology score If the value is higher than the threshold, the marine peptide sequence is considered and marine peptide sequences are redundant, delete one of them.

[0024] After data cleaning, if there are missing data in the marine peptide sequences, the sequence interpolation method is used to calculate the estimated value of the missing value through the interpolation formula.

[0025] In order to improve the accuracy of interpolation, a sequence interpolation method based on Lagrange interpolation was used. The There are missing values ​​at , the missing value is estimated by the position and value of the adjacent amino acids. Interpolation formula: , is the marine peptide sequence after data cleaning Middle The estimated value of amino acids is used to fill in the missing amino acid data; is the marine peptide sequence after data cleaning Middle The actual value of amino acids is the known amino acid data in the marine peptide sequence after data cleaning; Indicates the position index of the missing value, which is used to identify the position of the missing amino acid in the marine peptide sequence after data cleaning; Represents the temporary variable index used for interpolation calculation, which is used to traverse all the bit references involved in the interpolation process to ensure the calculation of the Lagrange interpolation formula; It is the number of adjacent amino acids, which determines the number of amino acids used in the interpolation calculation. It selects several non-missing amino acids before and after the missing value position. The specific value can be set according to the specific implementation scenario and is not limited here.

[0026] In order to unify the data format, the marine peptide sequence data after missing value filling is standardized into a unified format to obtain the preprocessed marine peptide sequence data. To facilitate subsequent processing, the specific standardization operation can be set according to the specific implementation scenario and is not limited here.

[0027] Furthermore, in order to improve the accuracy of marine peptide activity prediction, the pre-processed marine peptide sequence data is subjected to feature encoding and feature fusion processing to obtain a comprehensive feature set of the marine peptide sequence data. The specific process includes: based on the physicochemical properties of amino acids, the pre-processed marine peptide sequence data is subjected to topological sequence feature encoding, so as to convert the pre-processed marine peptide sequence data into a topological feature vector. Topological sequence feature encoding formula: , is the pre-processed marine peptide sequence The topological feature vector of Characteristics in topology; is the pre-processed marine peptide sequence Middle The weight of the amino acid topological features; is the exponential scaling factor of the topological feature, which is a constant and is used to adjust the influence of different topological features; is the pre-processed marine peptide sequence Middle The estimated value of amino acids; It is an amino acid The topological properties of Amino acids in marine peptide sequences after pretreatment The relative position and connection relationship in is the normalization coefficient used to adjust the scale of the logarithmic term; It is an amino acid Chemical structure functions, representing the pre-processed marine peptide sequences Middle The chemical properties of the amino acids, such as molecular size, polarity, etc. is the pre-processed marine peptide sequence The length of the preprocessed marine peptide sequence The total number of amino acids in.

[0028] In addition, the molecular orbital energy level feature encoding method is used to convert the preprocessed marine peptide sequence data into a molecular orbital energy level feature vector to represent the chemical substructure in the preprocessed marine peptide sequence. Molecular orbital energy level feature encoding formula: , is the pre-processed marine peptide sequence The molecular orbital energy level feature vector is used to represent the preprocessed marine peptide sequence Energy level information in molecular orbitals; is the pre-processed marine peptide sequence Middle The orbital energy level weights of each fragment; is the pre-processed marine peptide sequence Middle The orbital energy level value of each fragment indicates the position of the fragment on the quantum energy level; is the regularization parameter, which is a constant used to avoid the denominator being zero. is the adjustment coefficient, which is a constant used to control the size of the regularization term; is the pre-processed marine peptide sequence Middle fragments, which are used to split long sequences to fit the input of marine peptide activity prediction models; is the pre-processed marine peptide sequence The total number of fragments.

[0029] Furthermore, the topological feature vector and molecular orbital energy level feature vector of the preprocessed marine peptide sequence are fused using the topological-orbital feature fusion algorithm. The specific implementation formula is: , is a comprehensive feature vector, and the comprehensive feature set is composed of the comprehensive feature vector; and is the fusion weight of topological features and molecular orbital energy level features, indicating the relative importance of the two features in the fusion process; and is the pre-processed marine peptide sequence The topological eigenvector of The mean and standard deviation represent the average and dispersion of the topological features of all preprocessed marine peptide sequences; and is the pre-processed marine peptide sequence The molecular orbital energy level characteristic vector The mean and standard deviation represent the average and dispersion of the molecular orbital energy level characteristics of all preprocessed marine peptide sequences.

[0030] S2. Construct and train a marine peptide activity prediction model, and use the trained marine peptide activity prediction model to predict the activity of marine peptides based on a comprehensive feature set of marine peptide sequence data.

[0031] In order to predict the activity of marine peptides, the existing deep learning model is used to train the marine peptide activity prediction model, which specifically includes an input layer, a feature extraction layer, a feature fusion layer, a fully connected layer and an output layer; historical marine peptide sequence data is obtained through public databases and literature, and the input layer accepts the historical marine peptide sequence data preprocessed in step S1, and the feature extraction layer and the feature fusion layer perform feature processing, and the fully connected layer further processes the fused historical comprehensive feature set, and finally predicts the activity of marine peptides through the output layer. In the process of training the marine peptide activity prediction model, an adaptive dynamic feature optimization algorithm is introduced. The specific implementation process is as follows: First, initialize the marine peptide activity prediction model parameters. Set the following parameters: Marine peptide activity prediction model parameters , initial learning rate , first-order momentum parameter , second-order momentum parameter , small constant Parameters of the prediction model for marine peptide activity represents the connection strength in the marine peptide activity prediction model; initial learning rate Control the step size of each update of the marine peptide activity prediction model parameters; momentum parameter and Controls how much momentum affects the gradient estimate; small constant Prevent division by zero errors to ensure computational stability.

[0032] Calculating the loss function Parameters of the prediction model for marine peptide activity Gradient The loss function Reflects the difference between the predicted value and the true value of the marine peptide activity prediction model; Gradient Indicates the error direction of the current marine peptide activity prediction model. Calculation formula: , , is the loss function; is the parameter of the marine peptide activity prediction model, which indicates the connection strength in the marine peptide activity prediction model; is the number of samples, which represents the total number of samples of historical marine peptide sequence data; It is The true activity values ​​of historical marine peptide sequence data samples; Represents the marine peptide sequence data samples; The prediction model of marine peptide activity is based on The predicted activity value obtained for each sample is the output of the marine peptide activity prediction model; is the regularization parameter, which controls the weight of the regularization term to prevent overfitting; It is The absolute values ​​of the parameters of the marine peptide activity prediction model; is the total number of parameters of the marine peptide activity prediction model; is the regularization norm, which takes a value of 2 (L2 regularization).

[0033] Furthermore, the first-order momentum estimate is updated. The first-order momentum estimate represents the exponentially weighted average of the gradient, which smoothes the gradient fluctuation and enhances the stability of the marine peptide activity prediction model. Momentum estimate calculation formula: , is the first-order momentum Iteration estimation; is the first-order momentum parameter, which controls the influence of the first-order momentum in the gradient estimation and comes from the model hyperparameters; It is The first-order momentum estimate for the iteration.

[0034] Update the second-order momentum estimate. The second-order momentum estimate represents the exponentially weighted average of the squared gradient and helps adjust the learning rate changes. The specific formula is: , is the second-order momentum Iteration estimation; is the second-order momentum parameter, which controls the influence of the second-order momentum in the gradient estimation; It is The second-order momentum estimate for the iterations.

[0035] In order to eliminate the initial momentum deviation and improve the accuracy of the optimization process, the momentum correction is calculated and . Specific calculation formula: , , is the modified first-order momentum estimate; is the modified second-order momentum estimate.

[0036] Furthermore, the learning rate is adjusted dynamically; by combining the corrected first-order and second-order momentum estimates, the learning rate is adjusted dynamically , so that the learning rate can be adjusted dynamically to avoid the problem of too large or too small learning rate, and improve the convergence speed and prediction accuracy of the marine peptide activity prediction model. The formula for dynamically adjusting the learning rate is: , is the initial learning rate, is the dynamically adjusted learning rate.

[0037] Finally, update the parameters of the marine peptide activity prediction model . Through the dynamically adjusted learning rate and momentum estimate after momentum correction , , Update the parameters of the marine peptide activity prediction model , so that it is optimized in the direction of reducing the loss function. Optimization formula: , Indicates Model parameters after iterations; Indicates Model parameters after iterations; is a small constant used to avoid zero denominator.

[0038] Repeat the above steps until the model converges. Through continuous iteration, the parameters of the marine peptide activity prediction model are gradually optimized, and finally a trained marine peptide activity prediction model is obtained.

[0039] The trained marine peptide activity prediction model is used to predict the marine peptide activity based on the comprehensive feature set of the marine peptide sequence data obtained in step S1, thereby realizing the prediction of marine peptide activity based on deep learning.

[0040] In summary, a deep learning-based method for predicting the activity of marine peptides was completed.

[0041] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0042] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0043] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for predicting the activity of marine peptides based on deep learning, characterized in that: The following steps are involved: S1. Collect and preprocess marine peptide sequence data. The specific implementation process is as follows: The acquired marine peptide sequence data were cleaned. In the data cleaning stage, sequence homology scores were introduced and calculated. The redundancy was determined by calculating the homology scores between each pair of sequences. The calculation formula is: , It is a marine peptide sequence and marine peptide sequences Homology score; It is a marine peptide sequence The amino acids, It is a marine peptide sequence The amino acids; is the position index of the amino acid; is the indicator function; represents the total number of amino acids in the marine peptide sequence; After data cleaning is completed, if there are missing data in the marine peptide sequence, the sequence interpolation method is used to calculate the estimated value of the missing value through the interpolation formula. The interpolation formula is: , is the marine peptide sequence after data cleaning Middle The estimated value of amino acids; is the marine peptide sequence after data cleaning Middle The actual value of amino acids; Indicates the position index of the missing value; Indicates the temporary variable index for interpolation calculation; is the number of adjacent amino acids; The marine peptide sequence data after missing value filling were standardized into a unified format to obtain the preprocessed marine peptide sequence data; The preprocessed marine peptide sequence data is subjected to feature encoding and feature fusion processing to obtain a comprehensive feature set of the marine peptide sequence data. The specific implementation process is as follows: The preprocessed marine peptide sequence data is subjected to topological sequence feature encoding processing, and the preprocessed marine peptide sequence data is converted into a topological feature vector. The topological sequence feature encoding formula is: , is the pre-processed marine peptide sequence The topological eigenvector of ; is the pre-processed marine peptide sequence Middle The weight of the amino acid topological features; is the exponential scaling factor of the topological features; is the pre-processed marine peptide sequence Middle The estimated value of amino acids; It is an amino acid The topological properties of is the normalization coefficient; is the pre-processed marine peptide sequence Middle The chemical properties of the amino acids; Represents the pre-processed marine peptide sequence The total number of amino acids in The molecular orbital energy level feature coding process is used to convert the preprocessed marine peptide sequence data into a molecular orbital energy level feature vector. The molecular orbital energy level feature coding formula is: , is the pre-processed marine peptide sequence The molecular orbital energy level characteristic vector of ; is the pre-processed marine peptide sequence Middle The orbital energy level weights of each fragment; is the pre-processed marine peptide sequence Middle The orbital energy level values ​​of each fragment; is the regularization parameter; is the adjustment factor; is the pre-processed marine peptide sequence Middle fragments; is the pre-processed marine peptide sequence The total number of fragments; Based on the topological feature vectors of the preprocessed marine peptide sequences and the molecular orbital energy level feature vectors of the preprocessed marine peptide sequences, the topological-orbital feature fusion algorithm is used to perform feature fusion to obtain a comprehensive feature set of marine peptide sequence data. The specific implementation formula is: , is a comprehensive feature vector, and the comprehensive feature set is composed of the comprehensive feature vector; and It is the fusion weight of topological features and molecular orbital energy level features; and is the pre-processed marine peptide sequence The topological eigenvector of The mean and standard deviation of and is the pre-processed marine peptide sequence The molecular orbital energy level characteristic vector The mean and standard deviation of S2. Construct and train a marine peptide activity prediction model, and use the trained marine peptide activity prediction model to predict the activity of marine peptides based on a comprehensive feature set of marine peptide sequence data.

2. A method for predicting marine peptide activity based on deep learning according to claim 1, characterized in that: In said S2, it specifically includes: During the training process of the marine peptide activity prediction model, an adaptive dynamic feature optimization algorithm was introduced.

3. A method for predicting marine peptide activity based on deep learning according to claim 2, characterized in that: In the S2, it also includes: In the implementation process of the adaptive dynamic feature optimization algorithm, the model parameters of the marine peptide activity prediction model are initialized, including the marine peptide activity prediction model parameters, the initial learning rate, the first-order momentum parameter, the second-order momentum parameter and the constant used to prevent division by zero errors; and the gradient of the loss function to the model parameters of the marine peptide activity prediction model is calculated.

4. A method for predicting marine peptide activity based on deep learning according to claim 3, characterized in that: In the S2, it also includes: In the process of implementing the adaptive dynamic feature optimization algorithm, the first-order momentum estimation and the second-order momentum estimation are updated based on the gradient of the loss function to the model parameters of the marine peptide activity prediction model; and the first-order momentum estimation and the second-order momentum estimation are corrected.

5. A method for predicting marine peptide activity based on deep learning according to claim 4, characterized in that: In the S2, it also includes: In the implementation process of the adaptive dynamic feature optimization algorithm, the learning rate is dynamically adjusted based on the revised first-order and second-order momentum estimates; the model parameters of the marine peptide activity prediction model are updated through the dynamically adjusted learning rate and the revised first-order and second-order momentum estimates.

Images